@broberg/lens-daemon 0.1.8 → 0.1.10

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@broberg/lens-daemon",
-  "version": "0.1.8",
+  "version": "0.1.10",
   "description": "The Lens module of the cardmem daemon (flows, critic, coverage, capture sessions, the /lens/* routes and the cardmem-lens MCP) as a package. The daemon mounts createLensRouter(deps) + createLensMcpRouter(deps). Bun runtime.",
   "type": "module",
   "license": "MIT",
@@ -15,8 +15,8 @@
     "sharp": "^0.33.5",
     "turndown": "^7.2.4",
     "zod": "^3.23.8",
-    "@broberg/lens-svg": "0.1.0",
-    "@broberg/lens-engine": "0.11.0"
+    "@broberg/lens-engine": "0.12.0",
+    "@broberg/lens-svg": "0.1.0"
   },
   "devDependencies": {
     "@happy-dom/global-registrator": "^20.10.6",

package/src/flow.ts

@@ -50,6 +50,9 @@ export type ActivateMode = 'auto' | 'click' | 'enter';
  *  env var. The value is read at run time and never echoed. */
 export type ValueFrom = { path: string } | { env: string };
 
+/** F009.123 — a drag by an offset, in css pixels from the source's centre. */
+export type DragOffset = { x: number; y: number };
+
 export type FlowStep =
   // F074.51 — `timeout` was absent here while every other verb carried it, so a
   // goto's 30s was unoverridable. Found by the typecheck of this card's own
@@ -127,7 +130,11 @@ export type FlowStep =
   // for DnD features a synthetic-event assert can't cleanly drive (lens-gap 019f2438).
   | { type: 'clear'; testid?: string; target?: Target; label?: string; timeout?: number }
   | { type: 'press'; key: string; testid?: string; label?: string; timeout?: number }
-  | { type: 'drag'; testid: string; toTestid: string; label?: string; timeout?: number }
+  // F009.123 — `steps` mousemoves between press and release (default DRAG_STEPS): a
+  // d3-drag / React Flow node STARTS dragging on the first move and only moves on the
+  // next, so a one-hop drag passed while the node stayed put. `by` drags by an offset
+  // when the target is a canvas position, not an element. Exactly one of toTestid / by.
+  | { type: 'drag'; testid: string; toTestid?: string; by?: DragOffset; steps?: number; label?: string; timeout?: number }
   // F213.6 — uploadFile: set a fixture file on an input[type=file] (Playwright
   // setInputFiles), INCLUDING a hidden input behind a custom "add photo" button.
   // `path` is a repo-relative fixture path (sandboxed to local_path) or a data:
@@ -457,6 +464,10 @@ const SHARED_STEP_FIELDS = ['type', 'label', 'timeout', 'timeout_ms', 'browser']
  *  context (own cookies + storage), opened on first use and kept open for the rest
  *  of the run, so two people can be driven at the same time. `main` is the default. */
 const BROWSER_NAME = /^[a-z][a-z0-9_-]{0,23}$/;
+
+/** F009.123 — mousemoves per drag. jam measured 10 moving a React Flow node where 1
+ *  did not; more than one is what matters, ten keeps the travel smooth. */
+const DRAG_STEPS = 10;
 const TARGETED = ['testid', 'target'];
 export const STEP_FIELDS: Record<string, string[]> = {
   goto: ['path'],
@@ -483,7 +494,7 @@ export const STEP_FIELDS: Record<string, string[]> = {
   screenshot: [],
   clear: TARGETED,
   press: ['key', 'testid'],
-  drag: ['testid', 'toTestid'],
+  drag: ['testid', 'toTestid', 'by', 'steps'],
   uploadFile: ['testid', 'path'],
   autocomplete: ['testid', 'value', 'option', 'pickText', 'pickIndex'],
   capture: ['testid', 'as', 'from'],
@@ -618,8 +629,20 @@ function validateAndNormalizeSteps(steps: unknown[], depth = 0, inParallel = fal
       }
     }
     // F213.2 — drag needs a target testid; press needs a key (its testid is optional).
-    if (s.type === 'drag' && typeof (s as { toTestid?: unknown }).toTestid !== 'string') {
-      throw new Error('drag step requires a string toTestid');
+    if (s.type === 'drag') {
+      const d = s as { toTestid?: unknown; by?: unknown; steps?: unknown };
+      const hasTo = d.toTestid !== undefined;
+      const hasBy = d.by !== undefined;
+      if (hasTo === hasBy) throw new Error('drag step requires exactly one of toTestid (drag onto an element) or by {x,y} (drag by an offset)');
+      if (hasTo && typeof d.toTestid !== 'string') throw new Error('drag step requires a string toTestid');
+      if (hasBy) {
+        const b = d.by as { x?: unknown; y?: unknown } | null;
+        const keys = b && typeof b === 'object' && !Array.isArray(b) ? Object.keys(b).sort().join(',') : '';
+        if (keys !== 'x,y' || !Number.isFinite(b!.x) || !Number.isFinite(b!.y)) throw new Error('drag by must be {x: number, y: number} in css pixels');
+      }
+      if (d.steps !== undefined && !(Number.isInteger(d.steps) && (d.steps as number) >= 1 && (d.steps as number) <= 100)) {
+        throw new Error('drag steps must be a whole number from 1 to 100');
+      }
     }
     if (s.type === 'press' && typeof (s as { key?: unknown }).key !== 'string') {
       throw new Error('press step requires a string key');
@@ -964,9 +987,18 @@ function parseManuscriptStep(s: string): FlowStep {
     }
     case 'drag': {
       // "drag [source] -> [target]" (any two [testid]s in order).
+      // F009.123 — "drag [source] by dx,dy" for a canvas position, and an optional
+      // trailing "steps N" on either form.
       const ids = [...s.matchAll(/\[([^\]]+)\]/g)].map((m) => m[1]!.trim());
-      if (ids.length < 2) throw new Error(`drag step "${s}" needs [source] and [target] testids`);
-      return { type: 'drag', testid: ids[0]!, toTestid: ids[1]! };
+      const stepsM = s.match(/\bsteps\s+(\d+)\s*$/i);
+      const stepsPart = stepsM ? { steps: Number(stepsM[1]) } : {};
+      const byM = s.match(/\]\s+by\s+(-?\d+(?:\.\d+)?)\s*,\s*(-?\d+(?:\.\d+)?)/i);
+      if (byM) {
+        if (ids.length !== 1) throw new Error(`drag step "${s}" takes [source] by dx,dy OR [source] -> [target], not both`);
+        return { type: 'drag', testid: ids[0]!, by: { x: Number(byM[1]), y: Number(byM[2]) }, ...stepsPart };
+      }
+      if (ids.length < 2) throw new Error(`drag step "${s}" needs [source] and [target] testids, or [source] by dx,dy`);
+      return { type: 'drag', testid: ids[0]!, toTestid: ids[1]!, ...stepsPart };
     }
     case 'upload': {
       // "upload [input-testid] = <fixture>" — a repo-relative fixture path
@@ -1252,7 +1284,7 @@ function describe(step: FlowStep): string {
     case 'clickSelector': return `clickSelector ${step.selector}${step.at ? ` @ ${step.at}` : ''}`;
     case 'waitForUrl': return `waitForUrl ${step.url}`;
     case 'inspect': return step.html ? `inspect html${step.testid ? ` [${step.testid}]` : ''}` : step.frame ? 'inspect @frame' : 'inspect';
-    case 'drag': return `drag ${step.testid} → ${step.toTestid}`;
+    case 'drag': return `drag ${step.testid} ${step.by ? `by ${step.by.x},${step.by.y}` : `→ ${step.toTestid}`}`;
     case 'uploadFile': return `upload ${Array.isArray(step.path) ? `${step.path.length} files` : (step.path.startsWith('data:') ? 'data:…' : step.path)} → ${step.testid}`;
     case 'press': return `press ${step.key}${step.testid ? ` @${step.testid}` : ''}`;
     case 'autocomplete': return `autocomplete ${step.testid} → ${step.pickText !== undefined ? `"${step.pickText}"` : `#${step.pickIndex}`}`;
@@ -2100,14 +2132,28 @@ async function execStep(page: Page, step: FlowStep, base: string, vars: Record<s
         await page.keyboard.press(step.key);
       }
       return null;
-    case 'drag':
+    case 'drag': {
       // A real HTML5 drag-and-drop (dragstart→dragover→drop) via Playwright's
       // dragTo — drives the app's actual DnD handlers, unlike synthetic events.
-      await page
-        .locator(sel(step.testid))
-        .first()
-        .dragTo(page.locator(sel(step.toTestid)).first(), { timeout: stepTimeout(step) });
+      // F009.123 — in DRAG_STEPS mousemoves, not one: a d3-drag / React Flow node
+      // only starts dragging on the first move, so one hop left it in place, green.
+      const steps = step.steps ?? DRAG_STEPS;
+      const src = page.locator(sel(step.testid)).first();
+      if (step.by) {
+        await src.scrollIntoViewIfNeeded({ timeout: stepTimeout(step) });
+        const box = await src.boundingBox({ timeout: stepTimeout(step) });
+        if (!box) throw new Error(`drag: [data-testid="${step.testid}"] has no box to drag from`);
+        const x = box.x + box.width / 2;
+        const y = box.y + box.height / 2;
+        await page.mouse.move(x, y);
+        await page.mouse.down();
+        await page.mouse.move(x + step.by.x, y + step.by.y, { steps });
+        await page.mouse.up();
+        return null;
+      }
+      await src.dragTo(page.locator(sel(step.toTestid!)).first(), { timeout: stepTimeout(step), steps });
       return null;
+    }
     case 'uploadFile': {
       // F213.6 — set the fixture(s) on the target input[type=file] (works on a
       // HIDDEN input behind a custom button). setInputFiles takes either a list of

package/src/frames.ts

@@ -4,7 +4,9 @@
 // moment the probe already named, because a 600-frame contact sheet is a disk bill
 // and a token bill that answers nothing. The default is sparse-everywhere plus
 // dense-where-it-matters.
-import { spawn } from 'node:child_process';
+// F007.14 — ffmpeg is run and probed through the engine's ONE implementation, the same
+// one the fleet's video slicer (sliceVideo) uses; this file keeps only the fps cut.
+import { ffmpegAvailable, runFfmpeg } from '@broberg/lens-engine';
 import { mkdir, readdir, rm, stat } from 'node:fs/promises';
 import { join } from 'node:path';
 
@@ -31,14 +33,7 @@ export interface CutResult {
   unavailable?: string;
 }
 
-/** Is ffmpeg reachable at all? Answered once per call, cheaply. */
-export async function ffmpegAvailable(bin = 'ffmpeg'): Promise<boolean> {
-  return new Promise((resolve) => {
-    const p = spawn(bin, ['-version'], { stdio: 'ignore' });
-    p.on('error', () => resolve(false));
-    p.on('close', (code) => resolve(code === 0));
-  });
-}
+export { ffmpegAvailable };
 
 /**
  * Cut `video` into PNGs in `outDir`.
@@ -69,11 +64,7 @@ export async function cutFrames(
   // flag was doing nothing even on the versions that still accepted it.
   args.push('-vf', `fps=${fps}`, join(outDir, 'f-%05d.png'));
 
-  const code = await new Promise<number>((resolve) => {
-    const p = spawn(bin, args, { stdio: 'ignore' });
-    p.on('error', () => resolve(-1));
-    p.on('close', (c) => resolve(c ?? -1));
-  });
+  const code = await runFfmpeg(bin, args).then((r) => r.code, () => -1);
   if (code !== 0) {
     await rm(outDir, { recursive: true, force: true }).catch(() => {});
     return { frames: [], unavailable: `ffmpeg exited ${code} — video kept, no stills` };

package/src/mcp.ts

@@ -373,7 +373,7 @@ export const TOOLS = [
   {
     name: 'lens_run_manuscript',
     description:
-      `Compile a human-readable markdown manuscript (fmc-automation style: "# flow: name", "base: <url>", then one "- <verb> …" per line) into a flow and run it E2E. VERBS (${MANUSCRIPT_VERBS.length}, this is the complete list): ${MANUSCRIPT_VERBS.join(' / ')}. Shapes: "- goto /x" · "- fill [testid] = value" · a PASSWORD without plaintext: "- fill [testid] from file <path>" (absolute, or relative to local_path) or "- fill [testid] from env <NAME>" (F009.121 — password inputs only; the value never appears in the run, the report or a screenshot) · "- click [testid]" · "- press Enter" (global) or "- press Enter [testid]" (on a focused element) · "- expectText [testid] ~ text" · "- drag [source] -> [target]" · "- upload [input-testid] = <fixture>" · "- assert: <js>" · "- screenshot label" · frame-aware "- fillSelector <css> = <value> @frame <iframe>" and "- clickSelector <css> @frame <iframe>" · "- waitForUrl <substring>". TWO BROWSERS AT ONCE (F009.122): prefix a step with "in <name>:" — "- in b: goto /" · "- in b: expectText [note] == Hej" — and it runs in its own browser (own cookies + storage), open alongside the default one for the rest of the run; steps without the prefix run in the default browser. Click modifiers: "- click [testid] via enter" (F255.1, for Angular/MDC buttons a mouse click does not activate) and "- click [testid] @ 50%,28%" (F255.4, aim at a point inside a canvas/3D surface). A field that already HOLDS a value may no longer render an input — a combobox typically swaps to its selected state, so "fill" times out there while "press ArrowDown" / "press Enter" still drive it (fd-sundhed, 2026-08-21). The authoring shortcut — write prose, not JSON.`,
+      `Compile a human-readable markdown manuscript (fmc-automation style: "# flow: name", "base: <url>", then one "- <verb> …" per line) into a flow and run it E2E. VERBS (${MANUSCRIPT_VERBS.length}, this is the complete list): ${MANUSCRIPT_VERBS.join(' / ')}. Shapes: "- goto /x" · "- fill [testid] = value" · a PASSWORD without plaintext: "- fill [testid] from file <path>" (absolute, or relative to local_path) or "- fill [testid] from env <NAME>" (F009.121 — password inputs only; the value never appears in the run, the report or a screenshot) · "- click [testid]" · "- press Enter" (global) or "- press Enter [testid]" (on a focused element) · "- expectText [testid] ~ text" · "- drag [source] -> [target]" or "- drag [node] by 120,-40" (a canvas offset; both move in 10 mousemoves so React Flow / d3-drag nodes really move — "steps N" to change, F009.123) · "- upload [input-testid] = <fixture>" · "- assert: <js>" · "- screenshot label" · frame-aware "- fillSelector <css> = <value> @frame <iframe>" and "- clickSelector <css> @frame <iframe>" · "- waitForUrl <substring>". TWO BROWSERS AT ONCE (F009.122): prefix a step with "in <name>:" — "- in b: goto /" · "- in b: expectText [note] == Hej" — and it runs in its own browser (own cookies + storage), open alongside the default one for the rest of the run; steps without the prefix run in the default browser. Click modifiers: "- click [testid] via enter" (F255.1, for Angular/MDC buttons a mouse click does not activate) and "- click [testid] @ 50%,28%" (F255.4, aim at a point inside a canvas/3D surface). A field that already HOLDS a value may no longer render an input — a combobox typically swaps to its selected state, so "fill" times out there while "press ArrowDown" / "press Enter" still drive it (fd-sundhed, 2026-08-21). The authoring shortcut — write prose, not JSON.`,
     inputSchema: {
       type: 'object',
       required: ['project', 'manuscript'],